Object-Oriented Programming

PIMPL

Michael L. Collard, Ph.D.

Department of Computer Science, The University of Akron

Encapsulation & Information Hiding Rules

Place data and the operations that perform on that data in the same class

  • Don’t expose data items
  • Don’t expose the difference between stored data and derived data
  • Don’t expose implementation details of a class
  • Don’t expose a class’s internal structure

class Good

class Good

  • Actually is a “good class”
  • Follows encapsulation and information hiding rules
  • However the client code for Good Class:
  • Must #include "Good.hpp"
  • Knows data member types
  • Good.hpp may depend on a large number of direct and indirect include files for the data members

Client recompiled whenever change in class Good:

  • Add a data member
  • Delete a data member
  • Change the type of a data member
  • Update the include files for a data member
  • Even if there is no change in the interface

PIMPL

  • pointer to implementation
  • AKA, opaque pointer, d-pointer, compiler firewall, Cheshire Cat
  • Use a pointer to a class/struct declaration for an implementation class in the target class declaration file (i.e., .hpp)
  • Define the implementation class in the target class definition file (i.e., .cpp)

class Good PIMPL

PIMPL Advantages

  • The class implementation is hidden
  • Add a new data member to the (private) structure without violating binary compatibility
  • Header file containing the class declaration only needs to include those files required for the class interface, rather than for its implementation
  • Not a new technique archive_entry declaration, archive_entry definition

class Good PIMPL Implementation

PIMPL: Include File

  • Use a class or struct for the implementation. Typically a struct since no need to keep private.
  • A struct/class declaration forms an incomplete type
  • std::unique_ptr<> can work with with incomplete types (with one issue)
  • std::unique_ptr<> used instead of raw pointer for RAII. I.e., don’t depend on destructor of class to run

PIMPL: Implementation File

  • Define the PIMPL struct in the definition file
  • Target class data member initialization is now done in PIMPL struct
  • Access to “members” is through a pointer
  • If you have a default destructor, need to use the keyword delete with the destructor definition in the .cpp file so that std::unique_ptr<> can know how to delete the incomplete type.

class Stack PIMPL

A Step Further

A Step Further

  • We write most of the client code based on the abstract class Good
  • Often the abstract class is type we use as a parameter, constructor parameter, or field
  • Somewhere else in the code we choose a particular implementation
  • A Framework or Toolkit may be written using the abstract class Good
  • Then we provide the implementation class that the Framework or Toolkit uses

Choices

EndNotes

  • Abstractions can have a time efficiency cost
  • Abstractions can have a comprehension cost
  • Requirements for abstraction can vary over time on a project
  • Have to develop the skill to know when is enough